URVIS Workshop | CVPR 2026

Workshop Scope

URVIS investigates how unified perception can emerge from heterogeneous sensing. We explore architectures and training strategies that allow modalities to guide, compensate, or regularize one another under partial or missing signals.

Semantic Consistency

Aligned cross-modal features

RGB, depth, LiDAR, event, audio, tactile cues.

Unified Design

Sensor-aware models

Dynamic weighting by reliability and context.

Scaling Behavior

Quality vs. quantity

Few calibrated sensors vs. many redundant ones.

Cross-Modal Guidance

Fill gaps gracefully

Inference under dropouts, occlusions, or async data.

Core challenge: learning under missing or partial modalities while retaining robustness, interpretability, and efficiency for real-world deployment.

Call for Papers

We invite original research contributions on multisensor and multimodal perception, learning, and reasoning for robotics and embodied AI. The workshop aims to bring together researchers working on the integration, alignment, and exploitation of heterogeneous sensory signals for robust and intelligent robotic systems.

Submit on OpenReview

Topics of interest include, but are not limited to:

Multisensor and multimodal data fusion (early, late, hybrid, adaptive fusion)
Cross-modal alignment, correspondence, and synchronization
Multimodal representation learning and embedding spaces
Vision-based perception for robotics and autonomous systems
RGB-D, RGB-T, event-based, LiDAR, radar, audio, tactile, and proprioceptive sensing
Event cameras, neuromorphic sensing, and spiking neural networks
Multimodal foundation models and large-scale pretraining
Self-supervised, weakly supervised, and unsupervised multimodal learning
Multimodal tracking, detection, segmentation, and 3D understanding
Sensor calibration, registration, and geometric consistency
Uncertainty modeling, probabilistic fusion, and reliability-aware perception
Learning under missing, degraded, or noisy modalities
Cross-domain and cross-sensor generalization
Online, continual, and lifelong multimodal learning
Multimodal perception for manipulation, navigation, and human-robot interaction
Multimodal SLAM, mapping, and localization
Multimodal perception for autonomous driving and mobile robots
Efficient, real-time, and resource-aware multimodal systems
Simulation-to-real transfer and multimodal domain adaptation
Benchmark datasets, evaluation protocols, and reproducible research

We welcome regular paper submissions following the official CVPR Workshops submission pipeline. Accepted papers will be published in the CVPRW Proceedings and must comply with the standard CVPRW formatting guidelines.

All submissions will undergo a double-blind peer-review process to ensure fairness, rigor, and impartiality.

Challenges

Adverse-to-the-eXtreme Panoptic Segmentation

Multisensory (RGB, LiDAR, radar, event, IMU/GNSS) segmentation under clear and adverse conditions using MUSES data.

View details ->

Object Tracking in the Real World

Train with partial pairing; deploy with full or missing modalities.

View details ->

TBD

Challenge pending confirmation

Schedule

Time	Event
08:30	Welcome and Opening Remarks
08:40	Challenge Report 1
08:50	Challenge Report 2
09:00	Invited Talk 1
09:30	Invited Talk 2
10:00	Coffee Break
10:20	Invited Talk 3 (TBD)
10:30	Oral Presentation 1
10:40	Oral Presentation 2
10:50	Challenge Report 3
11:00	Invited Talk 4
11:30	Invited Talk 5
12:00	Awards and Closing Remarks